No one at Google or Facebook ever sat down and declared that stories about crime should take precedence over those highlighting achievement whenever a Black person appears in the news. That kind of directive was never committed to code by an engineer or approved as a specification by a product manager, and certainly no executive signed off on a memo bearing the title “Amplify Black Criminality.”
Yet the system produces exactly this result every hour of every day, across billions of news impressions served to hundreds of millions of users. The consistency would be impressive if it were intentional — but it is terrifying precisely because it is not.
Although the algorithm deciding what reaches you about Black people was never built to be racist, its goal of maximizing engagement produces identical outcomes — and that remains the central horror of the algorithmic age.
How the Machine Learns to Distort
Grasping this point begins with understanding what a recommendation algorithm does. Opening Google News or Facebook reveals not “the news” but a personalized selection of stories. Those stories were chosen by a machine learning model whose single goal is engagement — the probability you will click, read, share, or comment (Diakopoulos, Automating the News, Harvard University Press, 2019).
From billions of data points the model learned with ruthless efficiency that negative, threatening content gets more clicks than positive content. This is not new—the old newsroom saying “if it bleeds, it leads” predates the internet. The algorithmic era turned a human bias into a machine-optimized loop operating at a scale and speed no human process could match.
When a newspaper editor leads with a crime story, that choice affects only one edition. An algorithm prioritizing crime stories, however, generates millions of decisions every second, each reinforcing patterns that influence the next million. Over time the bias does not diminish; it compounds instead.
Engagement Disparity — Black Crime vs. Black Achievement Stories
Engagement data synthesis; Diakopoulos, Harvard University Press, 2019
The Overrepresentation Machine
In 2000 Travis Dixon and Daniel Linz published a study that should have transformed newsroom practices across America, yet it altered nothing. Their analysis of local television news in Los Angeles revealed Black people appearing as crime perpetrators far more frequently than their proportion of actual arrests warranted, while White people appeared less often as perpetrators and more often as victims (Dixon & Linz, Journal of Communication, 50(2), 2000).
The distortion was not subtle. It was consistent across stations and time periods. It was measurable with statistical precision.
Americans overestimate the proportion of crime committed by Black people by 20 to 30 percentage points. This is not a failure of individual perception. It is the predictable result of a media system that shows a reality where Black crime is wildly overrepresented.
Dixon and Linz documented in 2000 the human editorial version of that bias. Editors and producers followed the same engagement logic that algorithms would later automate, so their choices repeatedly overrepresented Black criminality. Human limits still constrained them. They could produce only so many broadcasts per day and remained, at least in theory, subject to professional norms.
The algorithm knows no such limits. Processing millions of stories daily, it runs nonstop without fatigue or conscience and has no concept of a Black person or of crime, only that certain words and images drive clicks. Those stories therefore spread wider, automating and amplifying the human editorial bias until it operates at a scale that makes local television news bias look small.
The Perception Distortion
Franklin Gilliam and Shanto Iyengar confirmed what Black viewers had long understood. Too much news coverage of Black crime leads people to inflate their estimates of Black offending. Study participants who saw a Black suspect became more inclined to back harsh crime policies and voiced more negative racial attitudes. Some even misremembered the suspect’s race (Gilliam & Iyengar, American Journal of Political Science, 44(3), 2000).
Staggering in scale, the distortion shows up clearly in surveys of news consumers. Americans overestimate the share of crime committed by Black people by 20 to 30 percentage points. Far from any failure of individual perception, the gap arises as the predictable result of a media system that wildly overrepresents Black crime. In the algorithmic era, this distortion has become industrialized.
Safiya Umoja Noble documented how search engines and recommendation systems reproduce and amplify racial stereotypes (Noble, Algorithms of Oppression, NYU Press, 2018). Searching for “Black girls” on Google in 2011 returned pornographic results at the top, while queries for “Black men” brought up content focused on criminality. No editorial decisions shaped those outcomes. Instead, the patterns emerged because the system had learned from millions of users what people wanted to see.
The algorithm did not create racism. It reflected racism, amplified it, and spread it so widely that it became part of the public systems of information itself.
What Does Your Real-World Intelligence Look Like?
The same data-driven rigor behind this article powers the Real World IQ assessment — measuring the cognitive ability that algorithms cannot distort.
Try 10 Free IQ Questions →The Feedback Loop That Shapes Policy
The consequences of algorithmic news bias go far beyond individual perception. They shape policy, elections, and the allocation of public resources.
- Distorted perception leads to distorted voting. Voters who believe Black crime rates are far higher than reality vote for candidates who promise to be “tough on crime” (Gilliam & Iyengar, 2000)
- Distorted voting leads to distorted policy. Legislators responsive to these voters allocate resources to policing over education. They fund incarceration over rehabilitation.
- Distorted policy leads to distorted outcomes. The resulting policies produce more arrests and more incarceration of Black people.
- Distorted outcomes feed more distortion. Those outcomes generate more crime stories. These stories generate more engagement. This trains the algorithm to distribute more crime stories.
“If you’re not careful, the newspapers will have you hating the people who are being oppressed, and loving the people who are doing the oppressing.”
— Malcolm X
This is the dangerous feedback loop. Rather than merely reflecting reality, the system shapes it first and then reflects that altered version, which molds the next round. Biased coverage creates biased algorithms. Those algorithms foster biased perceptions that shape policy, and the resulting outcomes generate still more biased coverage.
In systems theory this pattern is labeled a positive feedback loop. Lacking any natural end point, the cycle amplifies its own signal until the distortion becomes indistinguishable from reality.
The Strongest Counterargument — and Why the Data Defeats It
“The algorithm is neutral. It simply reflects what users want. If people click on crime stories more than achievement stories, that is a demand problem, not a supply problem. You cannot blame the mirror for the face.”
Three problems. First — The algorithm is not a mirror. A mirror shows you what is in front of it. An engagement-maximizing algorithm shows you what will make you click. It then reshapes its universe of content to produce more of it. It does not passively reflect demand. It actively manufactures it (Diakopoulos, Harvard, 2019). Second — The “users want it” defense ignores how the initial training data was generated. Human newsrooms were already biased toward Black crime stories before the algorithm existed (Dixon & Linz, 2000). The algorithm did not learn from neutral data. It learned from biased data and optimized the bias. Third — By the same logic, casinos are “neutral” because gamblers choose to gamble. The entire field of behavioral economics exists because humans are predictably irrational. Systems designed to exploit that irrationality bear responsibility for the exploitation.
The Newsroom Desert
Algorithmic bias operates against a backdrop of newsroom demographics that make editorial correction nearly impossible. According to the most recent data from the American Society of News Editors, Black journalists make up about 7% of newsroom staff at major outlets, a number that has barely moved in two decades. At the editorial decision-making level, the percentage is lower still.
Newsroom Demographics vs. U.S. Population
American Society of News Editors, Annual Survey
This matters because newsrooms make the human editorial decisions feeding the algorithm, yet they lack the perspectives needed to recognize bias. A newsroom that is 93% non-Black questions coverage of a Black crime story less often. A white crime story of equal severity might be ignored instead. Those who would pitch stories about Black achievement are not in the room.
The algorithm amplifies already-biased output from these unrepresentative newsrooms, building a distribution system that compounds the original bias at every stage.
Nicholas Diakopoulos documents the shift from human editorial judgment to algorithmic curation (Diakopoulos, Automating the News, Harvard University Press, 2019). When software decides which stories reach readers, the result is a system in which platform profits override journalistic values. A newspaper editor who repeatedly overrepresented Black criminality could face challenges from colleagues, yet an algorithm performing the same act stays protected as a trade secret, hidden by complexity and defended by companies that dismiss criticism as a misunderstanding of technology.
How Strong Is Your Relationship Intelligence?
The same analytical rigor behind this article powers the RELIQ assessment — measuring the emotional and relational intelligence that no algorithm can quantify.
Try 10 Free RELIQ Questions →The Puzzle and the Solution
How does a system with no conscious racial intent produce outcomes indistinguishable from a system designed to amplify Black criminality and suppress Black achievement — and how do you dismantle a bias that has no author?
A puzzle master examines the system to locate the variable generating the distortion, recognizing that the algorithm itself harbors no racism while the objective function does. The directive “Maximize engagement” becomes problematic when applied to a society already shaped by racial biases, yielding a machine that automates and scales those biases beyond any human capacity for manual correction.
Change the objective function. Mandate that engagement optimization be constrained by representational accuracy. Put the audit mechanism in public hands, not corporate ones.
Top 5 Solutions That Are Already Working
1. ProPublica Machine Bias (New York, covering Broward County, FL). ProPublica’s data journalism team revealed systemic racial bias in COMPAS criminal justice risk scores, finding that Black defendants were falsely flagged as high-risk at twice the rate of white defendants. The reporting prompted a Wisconsin court ruling that restricted the use of risk scores in sentencing. In this way ProPublica showed that algorithmic bias can be measured and challenged through investigative journalism (Angwin et al., ProPublica, May 2016; Buolamwini & Gebru, 2018).
2. Algorithmic Justice League (MIT Media Lab, Cambridge, MA). Joy Buolamwini founded the Algorithmic Justice League to audit commercial AI systems for racial and gender bias. The organization’s Gender Shades study revealed facial recognition disparities, including a 34.7% error rate for dark-skinned women compared with 0.8% for light-skinned men. Those results prompted IBM to leave the facial recognition market and led Amazon to place a moratorium on police use of its technology. AJL demonstrates how independent auditing can drive corporate accountability (Buolamwini & Gebru, Proceedings of ML Research, Vol 81, 2018).
3. Finland Media Literacy Curriculum (nationwide, all schools). From early childhood onward Finland makes media literacy a core skill in every school. Students practice critical evaluation of information and learn to resist disinformation. The country has ranked first in the European Media Literacy Index every year since 2017 and posted a score of 74 out of 100 in 2022. Among the 41 nations studied, Finland is the most resilient to disinformation. The record shows that citizens can be trained to recognize and resist algorithmic manipulation (Open Society Institute Sofia, 2023; Finnish National Agency for Education).
4. Capital B (Atlanta, GA and Gary, IN). Launched in 2022, Capital B operates as a Black-led nonprofit news organization that reports for Black communities through enterprise journalism and community listening. At launch the organization raised $9.4 million. Its reporting on hazardous Atlanta housing conditions produced direct repairs for affected residents. Capital B builds the counter-narrative algorithms suppress while delivering local, accountable journalism across the full spectrum of Black life (Nieman Journalism Lab, 2022; American Journalism Project).
5. Knight Foundation Press Forward (Miami, with nationwide grants). Press Forward represents a $500 million collaborative effort to rebuild local news public systems across America. Over five years the initiative has committed $300 million, of which $100 million has already been allocated. In 2024 alone it awarded more than 80 grants. More than 30 local Press Forward chapters now operate nationwide, and American Journalism Project partners have doubled in size through this funding. By rebuilding the local journalism public systems, Press Forward supplies the algorithm with stories beyond crime coverage (Knight Foundation, 2023–2024).
The Bottom Line
The numbers tell a story that no corporate deflection can override.
- 6X — The engagement disparity between Black crime stories and Black achievement stories (Diakopoulos, Harvard, 2019)
- 20–30 pts — The overestimation of Black crime share by American news consumers (Dixon & Linz, 2000)
- 7% vs. 13.6% — Black representation in newsrooms vs. the U.S. population (ASNE)
- Billions per day — The number of algorithmically curated news impressions that compound the bias every 24 hours
- Zero — The number of platforms that have voluntarily submitted to independent racial-disparity audits of their recommendation systems
The algorithm was not designed to be racist. Built instead to maximize engagement, it learned with inhuman efficiency that the most engaging stories about a Black person are those confirming the worst stereotypes. A click on a crime headline votes for more of the same. Shares function as training signals. Attention, second by second, teaches the machine to generate additional material matching what it has already decided you want.
The system is not broken. It is working perfectly, and that stands as the most dangerous sentence in this article. A broken system can be fixed, but one operating as designed requires redesign instead. The question is not whether the algorithm is biased but whether you are willing to stop feeding it.